How the data are built

What the numbers on the dashboard and in the briefs are, where they come from, and what they cannot tell you.

Sources

• Federal prosecutions. Press releases of the 94 U.S. Attorneys' Offices and the Department of Justice about child sexual exploitation cases, from 2013 onward, and published federal court opinions.
• State prosecutions. Press releases of the 50 state Attorneys General (or the state Department of Justice where that office prosecutes), and published state court opinions. Each week the newest releases are read and older ones are added, so the state archive grows over time. Where a state's own releases cannot be read automatically (currently Wisconsin and Wyoming), state cases are identified from news reports of the prosecution.
• Reported vs. prosecuted. The dashboard compares prosecutions with the CyberTipline reports that the National Center for Missing & Exploited Children (NCMEC) resolved to each state, from NCMEC's yearly "Reports by State" tables, with state populations from the Census Bureau's American Community Survey. A report is not a person or an offense: one report can concern a single file, and many reports can concern one person, so the ratio compares volumes and is not a prosecution rate. Reports are compared with prosecutions in the same or following years, because prosecutions take time. NCMEC publishes state totals only, not counties or federal districts.
• County prosecutors. Most state cases are prosecuted by county or district attorneys, who are not covered unless an Attorney General's office or a news report describes the case. State figures therefore undercount state prosecutions and should not be compared directly with federal counts.

What counts as a case

• A case is one person's prosecution. The releases about one defendant (charge, plea, sentence) are joined into one case when the name, the prosecuting office and the reported ages agree, allowing for the years between releases.
• Federal, state and dual jurisdiction. When the same person is prosecuted federally and by a state (same name and state, within six years, ages consistent), the two are one case labeled dual-jurisdiction. It is counted once in every total, appears under both the Federal and the State filter, and is the only kind shown under "Dual-jurisdiction only".
• Location on the map is the state of the prosecuting office or court, not where the offender or victims live.

Charge outcome

Compares the charging language in a case's releases with what the person pleaded guilty to or was convicted of:

• Convicted of CSAM offense: the plea or conviction names a CSAM offense (possession, receipt, distribution, production, promotion, or the state's equivalent), or the wording is too general to say otherwise.
• CSAM charge, convicted of non-CSAM offense: every plea or conviction stated names a different offense only, such as enticement, indecent exposure or sexual contact. A case is placed here only when the record states it explicitly.
• Charged, outcome not yet reported: a charge is described but no plea, verdict or sentence has been published yet.
• Not determined: the releases do not state the charge or the outcome in a recognizable form.

Categories and alerts

• Offender types follow Krone's (2004) typology; tactics, platforms, access, recruitment, monetization and other features are coded from the text of public records by keyword rules calibrated against hand-coded cases, and reviewed by hand where confidence is low. High-confidence codes are accepted without a manual review: a feature found in two or more releases about the case, a keyword rule whose agreement with the analyst's hand coding has a 95% confidence lower bound of at least 90%, or, where the official record is silent, a feature reported by news coverage matched to the defendant on several independent details (for example docket number, age and city). Where releases about the same case disagree, the broader category, then the majority of releases, then the latest release decides. Each week a random sample of automatically accepted codes is re-checked by hand, and a rule whose agreement falls below 90% returns to manual review. Treat category values as screens to verify before citing.
• Alerts compare the most recent years with the years before, with false-discovery-rate control. The word map shows terms in sentences describing offender conduct that rose or fell most; terms containing anyone's name are excluded.

Limits

• The data describe prosecuted and publicly documented cases. Changes reflect detection, enforcement priorities, charging practice and publicity as well as offending.
• No defendant names, victim information, images or media are held in the subscriber service or published.

Questions: support@northline-intelligence.com